Abstract
To encode a cognitive map of an environment, a navigating agent must be able to integrate across disparate perceptual views corresponding to the same place. There are two ways that this can be done. First, the agent might integrate across the panorama of views obtainable at a single vantage point. Second, they might integrate across views of a distal location containing a landmark that is visible from multiple vantage points. Guided by previous work, we tested the hypothesis that these two viewpoint-integration processes would be mediated by different neuroanatomical substrates. Male and female human participants were familiarized with a route through a virtual city, along which they closely viewed 24 storefronts that were pairwise associated, either by being located on different buildings directly across the street from each other (same-panorama condition) or by being located on different sides of the same building facing different streets (same-landmark condition). They were then scanned with fMRI while viewing the storefronts in isolation and performing a spatial memory task. Multivoxel pattern analysis revealed a functional distinction between two scene-selective regions: the retrosplenial complex (RSC) showed a significant association effect for same-panorama storefronts, whereas the parahippocampal place area (PPA) showed a significant association effect for same-landmark storefronts. Panoramic association effects were also observed in several other dorsal-stream regions within the medial and lateral parietal lobe. These results demonstrate the existence of two neural mechanisms for integrating across views to represent places as either the observer's location (same panorama) or the observed location (same landmark).